OpenAI’s GPT-Live, the new default voice model family for ChatGPT, launched on July 8, 2026, and introduces file attachments, project integration, memory, and search into voice sessions. GPT-Live is a family of full-duplex voice models that can listen and speak simultaneously. The flagship model is GPT-Live-1, with a lighter version called GPT-Live-1 mini for free-tier users. GPT-Live can delegate heavier cognitive tasks to more powerful models, including GPT-5.5. Paid subscribers get GPT-Live-1 with the full feature set, while free users access GPT-Live-1 mini. Users can attach files directly during a live voice conversation and have ChatGPT work with that content in real time. GPT-Live also supports image processing, memory features, and web search during voice sessions. It integrates with ChatGPT’s Projects feature, which maintains persistent context across sessions, including files, preferences, and custom instructions, spanning Work, Codex, and desktop environments. However, the ChatGPT Library, which stores documents separately from Projects, currently does not support direct file search or addition during a voice session. GPT-Live is an incremental maturation of OpenAI’s previously called Advanced Voice Mode, which introduced more natural prosody and emotional range but relied on a turn-based system. GPT-Live adds structural connections to the broader product ecosystem, including search, memory, file uploads, and Projects integration.